We used WRF-Chem model version 3.9.1 to simulate meteorological fields and O3 concentration in China. The model includes atmospheric physics and dynamical processes, atmospheric chemistry, and biophysical and biochemical processes (Grell et al., 2005; Skamarock and Klemp, 2008). The model domain is configured with 196 × 160 grid cells at 27 km horizontal resolution on the Lambert conformal projection and covers the entire mainland China. In the vertical direction, 28 layers are set extending from surface to 50 hPa. The meteorological initial and boundary conditions were adopted from ERA5 reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) at a horizontal resolution of 0.25° × 0.25° (Hersbach et al., 2020). The chemical initial and boundary conditions were generated from the Model for Ozone and Related Chemical Tracer version 4 (MOZART-4), which is available at a horizontal resolution of 1.9° × 2.5° with 56 vertical layers (Emmons et al., 2010).

Anthropogenic emissions are adopted from the 0.25° Multi-resolution Emission Inventory for China (MEIC) and MIX Asian emission inventory for the other regions (available at http://meicmodel.org, last access: 26 March 2024). Biogenic emissions are calculated online using the Model of Emissions of Gases and Aerosols from Nature (Guenther et al., 2006), which considers the impacts of plant types, weather conditions, and leaf area on vegetation emissions. Atmospheric chemistry is simulated using the Carbon Bond Mechanism version Z (CBM-Z) (Zaveri and Peters, 1999) gas-phase chemistry module coupled with a four-bin sectional Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) (Zaveri et al., 2008). The photolysis scheme is based on the Madronich fast-TUV photolysis module (Tie et al., 2003). The physical configurations include the Morrison double-moment microphysics scheme (Morrison et al., 2009), the Grell-3 cumulus scheme (Grell et al., 2002), the rapid radiative transfer model longwave radiation scheme (Mlawer et al., 1997), the Goddard shortwave radiation scheme (Chou and Suarez, 1994), the Yonsei University planetary boundary layer scheme (Hong et al., 2006), and the revised MM5 (Fifth-Generation Penn State/NCAR Mesoscale Model) Monin–Obukhov surface layer scheme.

Noah-MP is a land surface model coupled to WRF-Chem, with multiple options for key land–atmosphere interaction processes (Niu et al., 2011). Noah-MP considers the canopy structure with canopy height and crown radius and depicts leaves with prescribed dimensions, orientation, density, and radiometric properties. The model employs a two-stream radiative transfer approach for surface energy and water transfer processes (Dickinson, 1983). Noah-MP is capable of distinguishing photosynthesis pathways between C3 and C4 plants and defines vegetation-specific parameters for leaf photosynthesis and respiration.

Noah-MP considers prognostic vegetation growth through the coupling between photosynthesis and stomatal conductance (Farquhar et al., 1980; Ball et al., 1987). The photosynthesis rate, A (µmolCO2m-2s-1), is calculated as one of three limiting factors as follows: 1Atot=min⁡(WcWjWe)Igs, where Wc is the RuBisCo-limited photosynthesis rate, Wj is the light-limited photosynthesis rate, and We is the export-limited photosynthesis rate. Igs is the growing season index, with values ranging from 0 to 1. Stomatal conductance (gs) is computed based on the photosynthetic rate as follows: 2gs=1rs=mAnetCsRH+b, where b is the minimum stomatal conductance, m is the Ball–Berry slope of the conductance–photosynthesis relationship, Anet is the net photosynthesis by subtracting dark respiration from Atot, and Cs is the ambient CO2 concentration at the leaf surface. The assimilated carbon is allocated to various parts of vegetation (leaf, stem, wood, and root) and soil carbon pools (fast and slow), which determines the variations in the LAI and canopy height. Plant transpiration rate is then estimated using the dynamic LAI and stomatal conductance. Noah-MP also distinguishes the photosynthesis of sunlit and shaded leaves. Sunlit leaves are more limited by CO2 concentration, while shaded leaves are more constrained by insolation, leading to varied responses to O3 damage.

We implemented the O3 vegetation damage schemes proposed by Sitch et al. (2007) (hereafter S2007) and Lombardozzi et al. (2013) (hereafter L2013) into the Noah-MP. In the S2007 scheme, the undamaged fraction F for net photosynthesis is dependent on the sensitivity parameter aPFT and excessive area-based stomatal O3 flux, which is calculated as the difference between fO3 and threshold yPFT: 3F=1-aPFT×max⁡fO3-yPFT, where aPFT and yPFT are specifically determined for individual plant functional types (PFTs) based on measurements (Table 1). The stomatal O3 flux fO3 is calculated as 4fO3=[O3]ra+kO3×rs, where [O3] is the O3 concentration at the reference level (nmolm-3), and ra is the aerodynamic and boundary layer resistance between leaf surface and reference level (sm-1). kO3 = 1.67 represents the ratio of leaf resistance for O3 to that for water vapor. rs represents stomatal resistance (sm-1). For the S2007 scheme, stomatal conductance is damaged with the same ratio (1-F) as photosynthesis and further affects O3 uptake. In Noah-MP, the fO3 are calculated separately for sunlit and shaded leaves with corresponding stomatal resistance (Text S1 in the Supplement).

As a comparison, the L2013 scheme applies separate O3-damaging relationships for the photosynthetic rate and stomatal conductance. These independent relationships account for different plant groups and are calculated based on the cumulative uptake of O3 (CUO) under different levels of chronic O3 exposure. The leaf-level CUO (mmolm-2) is calculated by accumulating stomatal O3 fluxes of Eq. () from the start of the growing season to the specific time step with mean LAI > 0.5 (Lombardozzi et al., 2012), when vegetation is most vulnerable to air pollution episodes. O3 uptake is only accumulated when O3 flux is above an instantaneous threshold of 0.8 nmolO3m-2s-1 to account for ozone detoxification by vegetation at low O3 levels (Lombardozzi et al., 2015). We also include a leaf turnover rate for evergreen plants so that the accumulation of O3 flux does not last beyond the average foliar lifetime. The O3-damaging ratios depend on CUO, with empirical linear relationships as follows: 5FpO3=ap×CUO+bp6FcO3=ac×CUO+bc, where FpO3 and FcO3 are the ozone damage ratios for photosynthesis and stomatal conductance, respectively. The slopes (ap for photosynthesis and ac for stomatal conductance) and intercepts (bp for photosynthesis and bc for stomatal conductance) of regression functions are determined based on the meta-analysis of hundreds of measurements (Table 2). The ratios predicted in Eqs. () and () are applied to photosynthesis and stomatal conductance, respectively, to account for their independent responses to O3 damages. In Noah-MP, the FpO3 and FcO3 are calculated separately for sunlit and shaded leaves based on corresponding stomatal resistance (Text S1).

We validated the simulated meteorology and air pollutants with observations. The meteorological data were downloaded from the National Meteorological Information Center of the China Meteorological Administration (CMA Meteorological Data Centre, 2023; http://data.cma.cn/data/detail/dataCode/A.0012.0001.html, last access: 26 March 2024). The daily averaged surface pressure (PRES), wind speed at a height of 10 m (WS10), relative humidity (RH), and temperature at a height of 2 m (T2) were collected from 839 ground stations. Hourly surface O3 concentrations at 1597 sites in China were collected from Chinese National Environmental Monitoring Centre (CNEMC, https://air.cnemc.cn:18007/, last access: 26 March 2024).

We performed seven experiments to quantify the damaging effects of ambient O3 on GPP and the feedbacks to regional climate and air quality (Table 3). All simulations are conducted from 1 May to 31 August of 2017, with the first month excluded from the analysis as the spin-up. The control simulations (CRTL) excluded the impact of ozone on vegetation. Three offline simulations were performed with the same settings as the CTRL run, except that O3 vegetation damages were calculated and output without feedback to affect vegetation growth. These offline runs were established using either the S2007 scheme (Offline_SH07 for high sensitivity and Offline_SL07 for low sensitivity) or the L2013 scheme (Offline_L13). As a comparison, three online simulations applied the S2007 scheme (Online_SH07 for high sensitivity and Online_SL07 for low sensitivity) and the L2013 scheme (Online_L13) to estimate the O3 damages to GPP, which further influenced LAI development, leaf transpiration, and dry deposition. The differences between CTRL and online runs indicated the responses of surface meteorology and O3 concentrations to the O3-induced vegetation damages.

We compared the simulated summer near-surface temperature, relative humidity, wind speed, and surface O3 concentrations to observations. The model reasonably reproduces the spatial pattern of higher near-surface temperature in southeast and northwest and lower temperature over the Tibetan Plateau (Fig. 1a). On the national scale, the near-surface temperature is underestimated with a mean bias (MB) of 1.04 °C, but it shows a high correlation (R = 0.96). Unlike temperature, simulated relative humidity is overestimated with a MB of 5.04 % but a high R of 0.93 (Fig. 1b). Due to the modeling biases in the topographic effects, simulated wind speed is overestimated by more than 1.06 ms-1 on the national scale (Fig. 1c). Such an overestimation was also reported in other studies using WRF models (Hu et al., 2016; Liu and Wang, 2020; Zhu et al., 2022).

Comparisons with the measurements from air quality sites show that the simulated O3 deviates from the observed mean concentrations by 5.42 µgm-3 with a spatial R of 0.68. The model reasonably captures the hotspots over the North China Plain though with some overestimations, potentially attributed to uncertain emissions and coarse model resolutions. Such elevated bias in summer O3 is a common issue for both global and regional models over Asia. For example, Zhu et al. (2022) reported the overestimated summer average ozone concentration by 13.82 µgm-3 in China. Liu and Wang (2020) reached positive biases ranging from 3.7 µgm-3 to 13.32 µgm-3, using the WRF-CMAQ (Community Multiscale Air Quality) model. Overall, the WRF-Chem model shows reasonable performance in the simulation of surface meteorology and O3 concentrations in China.

We compared the offline O3 damage to photosynthesis between sunlit (PSNSUN) and shaded (PSNSHA) leaves during the summer. The S2007 scheme is dependent on instantaneous O3 uptake, which peaks in July when both O3 concentrations and stomatal conductance are high (Figs. S1 and S2 in the Supplement). For the same O3 pollution level, the damages are much higher for the sunlit leaves (Fig. 2a and b) than that for the shaded leaves (Fig. 2d and e) because of the higher stomatal conductance linked with the more active photosynthesis for the sunlit leaves. In contrast, the L2013 scheme depends on the accumulated O3 flux and assumes constant damages for some PFTs (Table 2), resulting in reductions in the photosynthesis even at low O3 concentrations. The O3 damage to photosynthesis of sunlit and shaded leaves increases month by month, reaching a maximum in August (Figs. S1 and S2). We found limited differences in the O3 damages between sunlit (Fig. 2c) and shaded (Fig. 2f) leaves with the L2013 scheme. Observations have reported that surface O3 has limited impacts on the shaded leaves (Wan et al., 2014), consistent with the results simulated by the S2007 scheme.

Figure 3 shows the effect of O3 damage to stomatal resistance of sunlit (RSSUN) and shaded (RSSHA) leaves. Overall, the spatial pattern of the changes in stomatal resistance is consistent with those of photosynthesis (Fig. 2) but with opposite signs. Both RSSUN and RSSHA are enhanced by O3 damage so as to prevent more O3 uptake. For the S2007 scheme, RSSUN with high and low sensitivities, respectively increases by 13.43 % (Fig. 3a) and 8.35 % (Fig. 3b), which are higher than the rates of 4.71 % (Fig. 3d) and 2.97 % (Fig. 3e) for RSSHA. These ratios are inversely connected to the changes in the photosynthesis (Fig. 2), suggesting the full coupling of damages between leaf photosynthesis and stomatal conductance. For the L2013 scheme, predicted changes in RSSUN (Fig. 3c) and RSSHA (Fig. 3f) are very similar with the magnitude of 25.3 %–26.3 %. These changes are higher than the loss of photosynthesis (Fig. 2c and f), suggesting the decoupling of O3 damages to leaf photosynthesis and stomatal conductance as revealed by the L2013 scheme.

We further assessed the O3 damage to GPP and transpiration (TR). For the S2007 scheme, O3 causes damages to national average GPP and TR approximately by 5.5 % with low sensitivity (Fig. 4b and e) and 8.4 % with high sensitivity (Fig. 4a and d) compared to the CTRL simulation. The model predicts high GPP damages over the North China Plain and moderate damages in the southeastern and northeastern regions. In the northwest, GPP damage is very limited due to the low relative humidity (Fig. 1b) that constrains the stomatal uptake. For the L2013 scheme, TR shows uniform reductions exceeding -25 % in most regions of China, except for the northwest (Fig. 4f), though O3 concentrations show a distinct spatial gradient (Fig. 1d). The changes in the GPP are similar to that of TR but with lower inhibitions (Fig. 4c). On average, the GPP reduction with the L2013 scheme is 2.5–3.9 times that predicted with the S2007 scheme. The most significant differences are located in Tibetan Plateau with limited damages in S2007 but strong inhibitions of both GPP and TR in L2013. The low temperature (Fig. 1a) and O3 concentrations (Fig. 1d) jointly constrain O3 stomatal uptake (Fig. S3 in the Supplement), leading to low O3 damages over Tibetan Plateau with the S2007 scheme. However, the L2013 scheme applies bp = 0.8021 for grassland (Table 2), suggesting strong baseline damages up to 20 %, even with CUO = 0 over Tibetan Plateau where the grassland dominates (Fig. S4 in the Supplement).

The O3 vegetation damage causes contrasting responses in surface sensible heat (SH) and LH (Fig. 5). For the S2007 scheme, the SH fluxes on average increase by 3.17 Wm-2 (8.85 %) with low sensitivity (Fig. 5b) and 5.99 Wm-2 (16.22 %) with high sensitivity (Fig. 5a). The maximum enhancement is located in southern China, where the increased stomatal resistance (Fig. 3a) reduces transpiration and the consequent heat dissipation. Meanwhile, LH fluxes decrease by 3.26 Wm-2 (5.58 %) with low sensitivity (Fig. 5e) and 6.43 Wm-2 (15.29 %) with high sensitivity (Fig. 5d), following the reduction in transpiration (Fig. 4d and e). We found similar changes in surface energy by O3–vegetation coupling between the S2007 and L2013 schemes. The SH shows the same hotspots over southern China with national average increase of 12.85 % (Fig. 5c), which is within the range of 8.85 % to 16.22 % predicted by the S2007 scheme. The LH largely decreases in central and northern China with the mean reduction of 17.4 % (Fig. 5f), which is close to the magnitude of 15.29 % predicted with the S2007 scheme using the high O3 sensitivity (Fig. 5d). Although the offline damages to GPP and TR are much larger with the L2013 than S2007 (Fig. 4), their feedback to surface energy shows consistent spatial pattern and magnitude (Fig. 5), likely because the O3 inhibition in S2007 has the same diurnal cycle with energy fluxes, while the L2013 scheme shows almost constant inhibitions throughout the day (Fig. S5 in the Supplement). The zero or near-zero slope parameters (ap and ac) in the L2013 scheme (Table 2) lead to insensitive responses of photosynthesis and stomatal conductance to the variations in CUO. As a result, there were very limited diurnal variations in O3 damage with the L2013 scheme. However, the strong nighttime damages in L2013 have limited contributions to the changes in surface energy, which usually peaks at the daytime.

The O3-induced damages to stomatal conductance weaken plant transpiration and thus slow down the heat dissipation at the surface, leading to the higher temperature but lower RH in China (Fig. 6). On the national scale, temperature increases by 0.5 °C due to O3 vegetation damage with the high sensitivity (Fig. 6a) and 0.23 °C with the low sensitivity (Fig. 6b) predicted using the S2007 scheme. A similar warming is predicted with the L2013 scheme, except that temperature shows moderate enhancement over Tibetan Plateau (Fig. 6c). The average RH decreases by 3.68 % with the high O3 sensitivity (Fig. 6d) and 2.22 % with the low sensitivity (Fig. 6e) in response to the suppressed plant transpiration. A stronger RH reduction of -3.85 % is achieved with the L2013 scheme, which predicts the maximum RH reductions in the north (Fig. 6f).

The O3-induced inhibition on stomatal resistance leads to a significant increase in surface O3 concentrations, particularly in eastern China (Fig. 7a–c). The main cause of such feedback is the reduction in O3 dry deposition, which exacerbates the O3 pollution in China. For the S2007 scheme, this positive feedback can reach up to 15 µgm-3 with high sensitivity (Fig. 7a) and 8 µgm-3 with low sensitivity (Fig. 7b) over the North China Plain. On the national scale, surface O3 enhances 4.40 µgm-3 (5.08 %) with high O3 sensitivity and 2.62 µgm-3 (3.04 %) with low O3 sensitivity through the coupling to vegetation. For the L2013 scheme, the changes in the O3 concentration (Fig. 7c) are comparable to that of the S2007 scheme with high sensitivity (Fig. 7a), except that the O3 enhancement is stronger in the southeast but weaker in the northeast.

The O3–vegetation coupling also increases surface isoprene emissions. For the S2007 scheme, isoprene emissions increase by 6.13 % with high sensitivity (Fig. 7d) and 3.43 % with low sensitivity (Fig. 7e), with regional hotspots in the North China Plain and northeastern and southern regions. The predictions using the L2013 scheme (Fig. 7f) show very similar patterns and magnitude of isoprene changes to the S2007 scheme with high sensitivity. Such an enhancement in isoprene emissions is related to the additional surface warming by O3–vegetation interactions (Fig. 6a–c). In turn, the increased isoprene emissions contribute to the deterioration of O3 pollution in China.